XCSP3 Competition 2017 Results · 2017. 8. 30. · (Java 8) Data (JSON) Compiler XCSP3 Instance...

transcript

XCSP3 Competition 2017– Results –

http://www.cril.fr/XCSP17/

presented by Christophe Lecoutre

23rd International ConferencePrinciples and Practice of Constraint Programming

August 30, 2017

XCSP3 is:

• an XML-based format designed to represent instances ofcombinatorial constrained problems

• an intermediate integrated format preserving the structure of themodels

XCSP3 is a major extension of XCSP 2.1 since it allows us to deal with:

• mono/multi optimization

• many types of variables

• cost functions

• reification and views

• annotatons

• variable quantification

• distributed, probabilistic and qualitative reasoning2

XCSP3: an Intermediate Format

ModelingLanguages

IntermediateFormat

FlatFormats

OPL, ESRA, MiniZ-inc, Essence, MCSP3, ...

XCSP 2.1, FlatZinc, wcsp

straction

www.xcsp.org

XCSP3: the central piece of a Modeling/Solving process

MCSP3 Model(Java 8)

Data(JSON)

Compiler

XCSP3 Instance(XML)

AbsCon Choco Mistral OscaR Sat4J ...

XCSP3: Available Tools and Benchmarks

Many tools are available on github:

https: // github. com/ xcsp3team/ .

Parsers available on github:

• Java 8 Parser

• C++ 11 Parser

Various tools for:

• checking solutions and bounds: org.xcsp.checker.SolutionChecker

• checking the validity of an instance for a competition track:org.xcsp.checker.CompetitionChecker

• checking the validity of an XCSP3 instance (made available soon)

Many series of CSP/COP instances that can be downloaded fromwww.xcsp.org by means of our selection engine!

Purpose of Competitions

The goal of a competition is to:

• evaluate solvers in the same conditions

• help collecting publicly available benchmarks and data (results,traces, . . . )

• help the community identify good ideas and strange results: the goalis to raise questions and get new ideas!

Competitions should not be misunderstood:

• The results are not an absolute truth: they depend on thebenchmark selection, experimental conditions, . . .

• A competition is not limited to a ranking: rankings are just anover-simplified view, but still relevant to motivate authors

• Competitions must be driven by the community: benchmarksubmission/selection advices, suggestions for improvements, . . .

Tracks for the 2017 XCSP3 Competition

There are 6 Standard tracks and two Minisolver tracks.

Problem Goal Exploration Timeout

CSP one solution sequential 40 minutesCSP one solution parallel 40 minutes

COP best solution sequential 4 minutesCOP best solution parallel 4 minutesCOP best solution sequential 40 minutesCOP best solution parallel 40 minutes

Table: Standard Tracks.

Problem Goal Exploration Timeout

CSP one solution sequential 40 minutesCOP best solution sequential 40 minutes

Table: Mini-Solver Tracks.

Perimeter of Constraints (mainly, XCSP3-core)

For the standard tracks:

• intension, extension

• regular and mdd

• allDifferent, allEqual, ordered and lex

• sum, count, nValues and cardinality

• maximum, minimum, element and channel

• noOverlap and cumulative

• instantiation

• slide

For the Mini-solver tracks:

• intension, extension

• allDifferent

• sum

• element

Computer Infrastructure

• The cluster we used is provided by CRIL and is composed of nodeswith two quad-cores (Intel @ 2.67GHz with 32 GiB RAM).

• Hyperthreading was disabled for the final runs.• Sequential solvers were run on one processor (4 cores) and were

allocated 15500 MiB of memory.• Parallel solvers were run on two processors (8 cores) and were

allocated 31000 MiB of memory.• The time limit can be understood either as a CPU limit, or as a

WCK (wall-clock) limit.• Sequential solvers are best compared with a CPU time limit.• If it is assumed that CPU cores come for free (which is quite a

strong assumption), both sequential and parallel solvers can beinterestingly compared with a WCK time limit.

Committees for the 2017 XCSP3 Competition

• OrganizationCRIL Christophe Lecoutre, Cedric Piette and Olivier RousselICTEAM Pierre SchaussI3S Arnaud MalapertLS2N Charles Prudhomme

• Judges• Claude-Guy Quimper from Universite Laval, Quebec, Canada• Helmut Simonis from Insight Centre for Data Analytics, Cork, Ireland• Christine Solnon from INSA, Lyon, France

Selection of Instances

After a few iterations, the jury has:

• decided how many instances should be selected in each availableseries of instances (from xcsp.org)

• chosen to randomly select instances in each series

For this first edition, we had to fix some problems (notably, the toolorg.xcsp.checker.CompetitionChecker was developped late).

Finally, the selection is as follows:

• Standard tracks: 510 CSP and 439 COP instances

• Mini-solver tracks: 242 CSP and 117 COP instances

Handling Submissions and Ranking

Olivier Roussel managed the submission of solvers.

• Bugged solvers were allowed to resubmit (up to several times, duringsummer).

• At any moment, no contestant had access to the selection of theinstances (managed by the jury and Olivier).

• Olivier has strong experience in the organisation of competitions(and is the developper of the competition infrastructure).

Ranking. based on the number of times a solver is able to give the bestknown answer (satisfiability, optimality, best known bound).

Teams/Solvers

In alphabetic order:

AbsCon-basic C. LecoutreBTD 2017-08-10 P. Jegou, H. Kanso, C TerriouxConcrete 3.4 J. VionMistral-2.0 E. Hebrard and M. SialaNaxos 1.1.0 N. PothitosOscaR ALNS, COS, Hybrid OscaR TeamOscaR - Parallel with EPS OscaR Teamchoco-solver 4.0.5 C. Prud’homme and J.-G. Fageschoco-solver 5a C. Prud’homme and J.-G. Fagescosoco (sat) 1.12 G. Audemardcosoco-mini 1.12 G. AudemardminiBTD P. Jegou, H. Kanso, C. Terriouxsat4j-CSP D. Le Berre, E. Lonca

CSP - sequential - CPU - 40 minutes (510 instances)

Rank Solver #solved SAT/UNSAT %inst. %VBS

Virtual Best Solver (VBS) 420 297 / 123 82% 100%

1 choco-solver 4.0.5 seq 372 264/108 73% 89%

2 choco-solver 5a 371 262/109 73% 88%

3 AbsCon-basic 368 266/102 72% 88%

4 Mistral-2.0 364 273/91 71% 87%

5 OscaR - Conflict Ord. 336 234/102 66% 80%

6 cosoco 1.12 335 238/97 66% 80%

7 Concrete 3.4 331 238/93 65% 79%

8 BTD 241 159/82 47% 57%

9 sat4j-CSP 235 168/67 46% 56%

1 choco-solver 4.0.5 seq 372 264/108 73% 89%

2 choco-solver 5a 371 262/109 73% 88%

3 AbsCon-basic 368 266/102 72% 88%

4 Mistral-2.0 364 273/91 71% 87%

6 cosoco 1.12 335 238/97 66% 80%

7 Concrete 3.4 331 238/93 65% 79%

8 BTD 241 159/82 47% 57%

9 sat4j-CSP 235 168/67 46% 56%

1 choco-solver 4.0.5 seq 372 264/108 73% 89%

2 choco-solver 5a 371 262/109 73% 88%

3 AbsCon-basic 368 266/102 72% 88%

4 Mistral-2.0 364 273/91 71% 87%

6 cosoco 1.12 335 238/97 66% 80%

7 Concrete 3.4 331 238/93 65% 79%

8 BTD 241 159/82 47% 57%

9 sat4j-CSP 235 168/67 46% 56%

1 choco-solver 4.0.5 seq 372 264/108 73% 89%

2 choco-solver 5a 371 262/109 73% 88%

3 AbsCon-basic 368 266/102 72% 88%

4 Mistral-2.0 364 273/91 71% 87%

6 cosoco 1.12 335 238/97 66% 80%

7 Concrete 3.4 331 238/93 65% 79%

8 BTD 241 159/82 47% 57%

9 sat4j-CSP 235 168/67 46% 56%

1 choco-solver 4.0.5 seq 372 264/108 73% 89%

2 choco-solver 5a 371 262/109 73% 88%

3 AbsCon-basic 368 266/102 72% 88%

4 Mistral-2.0 364 273/91 71% 87%

6 cosoco 1.12 335 238/97 66% 80%

7 Concrete 3.4 331 238/93 65% 79%

8 BTD 241 159/82 47% 57%

9 sat4j-CSP 235 168/67 46% 56%

COP - sequential - CPU - 40 minutes (438 instances)

Rank Solver #solved %inst. %VBS

Virtual Best Solver (VBS) 160 37% 100%

1 Mistral-2.0 133 30% 83%

2 OscaR - Conflict Ord. 125 29% 78%

3 AbsCon-basic 117 27% 73%

4 choco-solver 4.0.5 seq 115 26% 72%

5 OscaR - Hybrid 109 25% 68%

6 choco-solver 5a 106 24% 66%

7 Concrete 3.4 103 24% 64%

8 cosoco 1.12 98 22% 61%

9 cosoco-sat 1.12 98 22% 61%

10 sat4j-CSP 77 18% 48%

1 Mistral-2.0 133 30% 83%

4 choco-solver 4.0.5 seq 115 26% 72%

5 OscaR - Hybrid 109 25% 68%

7 Concrete 3.4 103 24% 64%

8 cosoco 1.12 98 22% 61%

9 cosoco-sat 1.12 98 22% 61%

10 sat4j-CSP 77 18% 48%

1 Mistral-2.0 133 30% 83%

4 choco-solver 4.0.5 seq 115 26% 72%

5 OscaR - Hybrid 109 25% 68%

7 Concrete 3.4 103 24% 64%

8 cosoco 1.12 98 22% 61%

9 cosoco-sat 1.12 98 22% 61%

10 sat4j-CSP 77 18% 48%

1 Mistral-2.0 133 30% 83%

4 choco-solver 4.0.5 seq 115 26% 72%

5 OscaR - Hybrid 109 25% 68%

7 Concrete 3.4 103 24% 64%

8 cosoco 1.12 98 22% 61%

9 cosoco-sat 1.12 98 22% 61%

10 sat4j-CSP 77 18% 48%

1 Mistral-2.0 133 30% 83%

4 choco-solver 4.0.5 seq 115 26% 72%

5 OscaR - Hybrid 109 25% 68%

7 Concrete 3.4 103 24% 64%

8 cosoco 1.12 98 22% 61%

9 cosoco-sat 1.12 98 22% 61%

10 sat4j-CSP 77 18% 48%

CSP - seq. and // - WCK - 40 minutes (510 instances)

1 choco-solver 4.0.5 // 395 286/109 77% 94%

2 choco-solver 4.0.5 seq 372 264/108 73% 89%

3 choco-solver 5a 371 262/109 73% 88%

4 AbsCon-basic 368 266/102 72% 88%

5 Mistral-2.0 364 273/91 71% 87%

7 cosoco 1.12 335 238/97 66% 80%

8 Concrete 3.4 331 238/93 65% 79%

9 OscaR - // with EPS 327 226/101 64% 78%

10 BTD 241 159/82 47% 57%

11 sat4j-CSP 235 168/67 46% 56%

1 choco-solver 4.0.5 // 395 286/109 77% 94%

2 choco-solver 4.0.5 seq 372 264/108 73% 89%

3 choco-solver 5a 371 262/109 73% 88%

4 AbsCon-basic 368 266/102 72% 88%

5 Mistral-2.0 364 273/91 71% 87%

7 cosoco 1.12 335 238/97 66% 80%

8 Concrete 3.4 331 238/93 65% 79%

9 OscaR - // with EPS 327 226/101 64% 78%

10 BTD 241 159/82 47% 57%

11 sat4j-CSP 235 168/67 46% 56%

1 choco-solver 4.0.5 // 395 286/109 77% 94%

2 choco-solver 4.0.5 seq 372 264/108 73% 89%

3 choco-solver 5a 371 262/109 73% 88%

4 AbsCon-basic 368 266/102 72% 88%

5 Mistral-2.0 364 273/91 71% 87%

7 cosoco 1.12 335 238/97 66% 80%

8 Concrete 3.4 331 238/93 65% 79%

9 OscaR - // with EPS 327 226/101 64% 78%

10 BTD 241 159/82 47% 57%

11 sat4j-CSP 235 168/67 46% 56%

1 choco-solver 4.0.5 // 395 286/109 77% 94%

2 choco-solver 4.0.5 seq 372 264/108 73% 89%

3 choco-solver 5a 371 262/109 73% 88%

4 AbsCon-basic 368 266/102 72% 88%

5 Mistral-2.0 364 273/91 71% 87%

7 cosoco 1.12 335 238/97 66% 80%

8 Concrete 3.4 331 238/93 65% 79%

9 OscaR - // with EPS 327 226/101 64% 78%

10 BTD 241 159/82 47% 57%

11 sat4j-CSP 235 168/67 46% 56%

1 choco-solver 4.0.5 // 395 286/109 77% 94%

2 choco-solver 4.0.5 seq 372 264/108 73% 89%

3 choco-solver 5a 371 262/109 73% 88%

4 AbsCon-basic 368 266/102 72% 88%

5 Mistral-2.0 364 273/91 71% 87%

7 cosoco 1.12 335 238/97 66% 80%

8 Concrete 3.4 331 238/93 65% 79%

9 OscaR - // with EPS 327 226/101 64% 78%

10 BTD 241 159/82 47% 57%

11 sat4j-CSP 235 168/67 46% 56%

COP - seq. and // - WCK - 40 minutes (438 instances)

1 choco-solver 4.0.5 // 145 33% 91%

2 OscaR - // with EPS 135 31% 84%

3 Mistral-2.0 133 30% 83%

6 choco-solver 4.0.5 seq 115 26% 72%

7 OscaR - Hybrid 109 25% 68%

9 Concrete 3.4 103 24% 64%

10 cosoco 1.12 98 22% 61%

11 cosoco-sat 1.12 98 22% 61%

12 sat4j-CSP 77 18% 48%17

1 choco-solver 4.0.5 // 145 33% 91%

2 OscaR - // with EPS 135 31% 84%

3 Mistral-2.0 133 30% 83%

6 choco-solver 4.0.5 seq 115 26% 72%

7 OscaR - Hybrid 109 25% 68%

9 Concrete 3.4 103 24% 64%

10 cosoco 1.12 98 22% 61%

11 cosoco-sat 1.12 98 22% 61%

12 sat4j-CSP 77 18% 48%17

1 choco-solver 4.0.5 // 145 33% 91%

2 OscaR - // with EPS 135 31% 84%

3 Mistral-2.0 133 30% 83%

6 choco-solver 4.0.5 seq 115 26% 72%

7 OscaR - Hybrid 109 25% 68%

9 Concrete 3.4 103 24% 64%

10 cosoco 1.12 98 22% 61%

11 cosoco-sat 1.12 98 22% 61%

12 sat4j-CSP 77 18% 48%17

1 choco-solver 4.0.5 // 145 33% 91%

2 OscaR - // with EPS 135 31% 84%

3 Mistral-2.0 133 30% 83%

6 choco-solver 4.0.5 seq 115 26% 72%

7 OscaR - Hybrid 109 25% 68%

9 Concrete 3.4 103 24% 64%

10 cosoco 1.12 98 22% 61%

11 cosoco-sat 1.12 98 22% 61%

12 sat4j-CSP 77 18% 48%17

1 choco-solver 4.0.5 // 145 33% 91%

2 OscaR - // with EPS 135 31% 84%

3 Mistral-2.0 133 30% 83%

6 choco-solver 4.0.5 seq 115 26% 72%

7 OscaR - Hybrid 109 25% 68%

9 Concrete 3.4 103 24% 64%

10 cosoco 1.12 98 22% 61%

11 cosoco-sat 1.12 98 22% 61%

12 sat4j-CSP 77 18% 48%17

2 Mistral-2.0 115 26% 80%

3 choco-solver 4.0.5 seq 103 24% 72%

5 OscaR - Hybrid 100 23% 69%

7 Concrete 3.4 90 21% 63%

8 cosoco-sat 1.12 87 20% 60%

9 cosoco 1.12 85 19% 59%

10 sat4j-CSP 60 14% 42%

2 Mistral-2.0 115 26% 80%

3 choco-solver 4.0.5 seq 103 24% 72%

5 OscaR - Hybrid 100 23% 69%

7 Concrete 3.4 90 21% 63%

8 cosoco-sat 1.12 87 20% 60%

9 cosoco 1.12 85 19% 59%

10 sat4j-CSP 60 14% 42%

2 Mistral-2.0 115 26% 80%

3 choco-solver 4.0.5 seq 103 24% 72%

5 OscaR - Hybrid 100 23% 69%

7 Concrete 3.4 90 21% 63%

8 cosoco-sat 1.12 87 20% 60%

9 cosoco 1.12 85 19% 59%

10 sat4j-CSP 60 14% 42%

2 Mistral-2.0 115 26% 80%

3 choco-solver 4.0.5 seq 103 24% 72%

5 OscaR - Hybrid 100 23% 69%

7 Concrete 3.4 90 21% 63%

8 cosoco-sat 1.12 87 20% 60%

9 cosoco 1.12 85 19% 59%

10 sat4j-CSP 60 14% 42%

2 Mistral-2.0 115 26% 80%

3 choco-solver 4.0.5 seq 103 24% 72%

5 OscaR - Hybrid 100 23% 69%

7 Concrete 3.4 90 21% 63%

8 cosoco-sat 1.12 87 20% 60%

9 cosoco 1.12 85 19% 59%

10 sat4j-CSP 60 14% 42%

Mini-solver Tracks - sequential - CPU - 40 minutes

Results for CSP (242 instances)

1 cosoco-mini 1.12 181 119/62 75% 95%